where h(t, xi; λ, w) = wδ(λxi≤ t) + (1 − w)δ(λxi> t) and w = {0, 1}. Note that w = 0 could be a good choice if therelationship between y and x is decreasing and w = 1 could be a good choice if such relationship is increasing. Notethat F∗rw(t) also depends on w and λ, but we use F∗rw(t) instead of F∗rw(t, w, λ) to achieve a simpler notation.The proposed class of ratio estimators can also be motivated along the lines of [14], i.e., using standard results fortotals or means to define a ratio estimator. In particular, the class of ratio estimators can be expressed as